PopAnesQA: leaderboard
PopAnesQA (Shin et al., MLHC 2026, PMLR 340): 623 four-option, scenario-based multiple-choice anesthesiology questions, each built around a patient profile. Unlike MedQA-style knowledge questions, it asks whether a model can make "action-oriented clinical management decisions that are correctly adapted to the target patient population"; the authors found pediatric anesthesia to be "only 0.2%" of five widely used medical QA sets. Answered without retrieval at temperature 0. Every item was generated with Gemini-3-Flash, which also scores highest. Score: accuracy (%) over all 623 questions. Higher is better. Source: https://proceedings.mlr.press/v340/shin26a.html
Metric: Accuracy (%). Source: proceedings.mlr.press. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 87 | #103 |
| 2 | GPT-4o | 74 | #357 |
| 3 | Gemma 3 27B (IT) | 69.66 | #528 |
| 4 | Mixtral 8x7B Instruct | 64.21 | #1085 |
| 5 | Llama 3.1 8B Instruct | 60.83 | #1017 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=popanesqa · How It Works · Data refreshed daily, snapshot 2026-11-05.